links between single-trial changes and learning rate in
TRANSCRIPT
ORIGINAL PAPER
Links Between Single-Trial Changes and Learning Rate in EyelidConditioning
Andrei Khilkevich1& Hunter E. Halverson1
& Jose Ernesto Canton-Josh1&
Michael D. Mauk1,2
# Springer Science+Business Media New York 2015
Abstract The discovery of single-trial learning effects, wherethe presence or absence (or the number) of climbing fiberinputs produces measureable changes in Purkinje cell re-sponse and in behavior, represents a major breakthrough incerebellar learning. Among other things, these observationsprovide strong links between climbing fiber-mediated plastic-ity and cerebellar learning. They also demonstrate that cere-bellar learning is stochastic, with each instantiation of a move-ment producing a small increment or decrement in gain. Thesum of the small changes give rise to the macroscopic prop-erties of cerebellar learning. We used a relatively large data setfrom another example of cerebellar-dependent learning, clas-sical conditioning of eyelid responses, to attempt a behavioralreplication and extension of single-trial learning effects. As anormal part of training, stimulus-alone trials provide instanceswhere the climbing fiber response would be omitted, similarto non-climbing-fiber trials (gain down) during smooth pur-suit training. The consequences of the stimulus-alone trial onthe amplitude and timing of the conditioned response on thefollowing paired trials were examined. We find that the am-plitude of the conditioned response during the trial after astimulus-alone trial (no climbing fiber input) was measurablysmaller than the amplitude on the previous trials, and thissingle-trial effect on amplitude is larger for longer interstimu-lus intervals. The magnitude of the single-trial effect parallelsthe rate of extinction at different interstimulus intervals
supporting the previously observed link between single-trialeffects and learning.
Keywords Climbing fiber . LTD . LTP . Stochastic learning .
Timing
Introduction
A strong understanding of cerebellar learning requires estab-lishing causal links between plasticity at cerebellar synapsesand changes in behavior mediated by the cerebellum. It is alsoimportant to determine how properties of plasticity interact withproperties of circuits to confer to the cerebellum its processingand learning capabilities. A series of studies from the Lisbergerlab [1, 2] provided a remarkable step toward these goals usingsmooth-pursuit eye movements. They demonstrated that thepresence or absence of a climbing fiber input to a Purkinje cellproduces measurable changes in both the Purkinje cell responseand the pursuit response on the next trial. Because the directionof these changes is consistent with the climbing-fiber-inducedplasticity at cerebellar synapses, these experiments represent asstrong a correlation as can be established between plasticity atPurkinje cell synapses and cerebellar learning. These data alsodemonstrate that cerebellar learning is incremental and to adegree stochastic. Each movement will either involve aclimbing fiber input or not, and thus the next movement shouldeither be a little larger or smaller, respectively.
The fundamental importance of these findings highlightsthe value of replication using a different cerebellar-dependentform of learning such as eyelid conditioning. Particular prop-erties of eyelid conditioning permit a relatively powerful yetsimple to implement test of single-trial changes in cerebellarlearning. Eyelid conditioning involves pairing a conditionedstimulus (CS) such as a tone with a reinforcing unconditioned
* Michael D. [email protected]
1 Center for Learning andMemory, The University of Texas at Austin,1 University Station C7000, Austin, TX 78712-0805, USA
2 Department of Neuroscience, The University of Texas at Austin,Austin, TX 78712, USA
CerebellumDOI 10.1007/s12311-015-0690-8
stimulus (US), which in our experiments involves subdermalstimulation of the skin near the eye. In untrained animals, theUS elicits a reflex response—the eyelids close. With pairedpresentations of CS and US, the tone CS acquires the ability toelicit a learned eyelid closure. As outlined in numerous re-views, what makes this learning useful is that the CS is con-veyed to the cerebellum via activation of mossy fibers and theUS by activation of climbing fiber inputs to the cerebellum [3,4]. Output from the anterior interpositus deep nucleus drivesthe expression of conditioned responses (CRs) [5–7]. Eyelidconditioning therefore provides an experimentally tractablemeans to control cerebellar inputs and to infer cerebellar out-put via measurement of eyelid responses. In order to gainmore control over the precise mossy fiber inputs to the cere-bellum relative to a tone stimulus, it is possible to directlyelectrically stimulate mossy fibers as a substitute for the pe-ripheral CS. This technique isolates learning mechanisms inthe cerebellum from outside influences and produces learningthat is indistinguishable from learning with a tone CS [8, 9].
Here, we make use of a common feature of eyelid condi-tioning training to implement a test of single-trial changes inbehavior mediated by the cerebellum. Most training sessionsinvolve occasional CS-alone trials to reveal the full timecourse of the CRs uncontaminated by the reflexive responseto the US. When CS-alone trials are given repeatedly, extinc-tion occurs, where response amplitude decreases until, withenough CS-alone training, the responses can fully disappear[10]. Previous studies have demonstrated that climbing fiberinputs to Purkinje cells are suppressed during the expressionof a conditioned response (CR) on CS-alone trials [11, 12] andthat indeed this suppression is the signal for extinction (or gaindown) learning in the cerebellum [13]. This means that inwell-trained animals, these occasional CS-alone trials repre-sent instances where the CS was presented almost certainly inthe absence of a climbing fiber input.
We made use of the very large number of animals that havebeen trained in the delay eyelid conditioning paradigm to askwhether there is a detectable decrease in CR amplitude in thetrial following the CS-alone trials. While this is a purely be-havioral—and one-directional—attempt at replicating theMe-dina and Lisberger [1] findings, the large number of animalsthat have been trained this way provides strong statisticalpower to investigate the details of the behavioral effects ofCS-alone trials on CRs. In addition, because different groupsof animals were conditioned with either tone or electricalmossy fiber stimulation as a CS, comparing single-trial effectsbetween these groups, we can either detect or exclude contri-butions of upstream CS processing from specifically cerebel-lar learning mechanisms. Moreover, for both CSs, animalswere trained with a broad range of inter-stimulus intervals(ISIs), ranging from ISI 200 to ISI 1500, thus permitting in-vestigation of the interaction between single-trial effects withISI length. We have observed, for example, an almost 20-fold
difference in the rate of extinction of CRs between the shortestand longest ISIs examined in the present analyses (Ohyamaand Mauk, unpublished). Whereas, at the shortest ISI, re-sponses may require several 100-trial sessions to extinguish,responses from animals trained at the longer ISIs can requireas few as 10 trials to extinguish. Thus, we hoped to see wheth-er there are ISI-dependent differences in the changes in CRsthat immediately follow CS-alone trials.
Our results provide a clear replication of the main single-trial effect first shown by Medina and Lisberger and furtherdemonstrate that the size of the changes as a consequence ofsingle CS-alone trials scales with ISI duration and matches thedifferent rates of extinction for those ISIs. The data also clear-ly demonstrate that quite similar results are observed indepen-dent of whether training involves a tone CS or stimulation ofmossy fibers, and thus that any factors upstream of the cere-bellum can be excluded. These findings support the implica-tions of the original single-trial findings by showing that sin-gle trials with no climbing fiber input produce detectable de-creases in responses on the next trial. The present data are alsoclearly consistent with the incremental and stochastic natureof cerebellar learning first revealed by Medina and Lisberger.
Methods
Subjects
Data were obtained from 160 male New Zealand albino rab-bits (Oryctolagus cuniculus, Myrtle’s Rabbitry, ThompsonsStation, TN). The animals weighed between 2.5 and 3 kg atthe time of surgery, were individually housed, were fed daily,and had free access to water. Treatment of animals and surgi-cal procedures were in accordance with National Institutes ofHealth Guidelines and an institutionally approved animal wel-fare protocol.
Surgery
Regardless of experimental preparation, all rabbits received asimilar surgery with respect to the method of behavioral dataacquisition. Before surgery, a preanesthetic (40 mg/kg keta-mine and 5 mg/kg acepromazine) was injected subdermally,and each rabbit was positioned in a stereotaxic restrainer suchthat lambdawas 1.5 mmventral to bregma. General anesthesiawas maintained with isofluorene (2 % mixed in oxygen), andsterile procedures were used throughout the surgery. After alidocane injection into the scalp, a midline incision was madeand the skin and underlying tissue were retracted and held inplace with hemostats. Three holes were drilled in the skull,which allowed insertion of stainless steel anchor screws. An-chor screws also functioned as ground screws in the mossyfiber stimulation animals. Rabbits prepared for mossy fiber
Cerebellum
stimulation received a single or two laterally spaced (1 mm)stimulating electrodes (A-M Systems, Carlsborg, WA; tip ex-posed to obtain impedance of 100–200 kΩ) in the middlecerebellar peduncle ipsilateral to the trained eye (5.5 mm lat-eral, 16 mm ventral, and 3 mm anterior from lambda). Theanchor/ground screws, mossy fiber stimulation electrodes,and a head bolt for the infrared emitter/detector were all se-cured to the skull with dental acrylic. Stainless steel loopsterminating in gold pins were inserted into the caudal androstral periorbital region of the left eye for delivery of thestimulation US. Rabbits were given postoperative analgesicsand antibiotics for 2 days after surgery and were allowed torecover for a week before experiments began.
Conditioning
Animals were trained in custom-designed, well-ventilated,and sound-attenuating chambers measuring 90×60×60 cm(length, width, height). Each chamber was equipped with aspeaker that was connected to an audio source module (modelV85-05, Coulborn Instruments, Allentown, PA) or Windows-based PC used to generate tones. Electrical leads from a stim-ulus isolator (model 2100, A-M Systems, Carlsborg, WA)were attached to the periorbital electrodes to deliver pulsesof electrical stimulation used as the US. For rabbits trainedwith mossy fiber stimulation as a CS, separate stimulatorswere used to time the delivery of constant current pulses ini-tiated by custom software through stimulus isolators (model2300, A-M Systems, Carlsborg, WA), which were connectedwith gold pins to the electrodes implanted in the middle cere-bellar peduncle. To measure eyelid position, an infraredemitter/detector was attached directly to the head stage of eachrabbit to record movements of the left external eyelid. Thesedetectors provide a linear readout of eyelid position by mea-suring the amount of infrared light reflected back to the detec-tor, which increases as the eye closes [14]. At the start of eachdaily session, the detector was calibrated by delivering the USto elicit maximum eye closure. The amplification of the signalwas adjusted to match an assumed maximum eye closure of6 mm.
Stimulus presentation was controlled by custom-designedsoftware for all experiments. Rabbits were given daily eyelidconditioning sessions comprised of 12 blocks of 9 trials (1 CS-alone trial and 8 paired trials per block, 108 trials total). Insome cases, the CS-alone trial occurred at the beginning ofeach block while other training paradigms presented the CS-alone trial at the end of each block. Training trials involvingpresentation of a tone CS were typically a 1- or 9.5-kHz, 85-dB sinusoidal tone with a rise and fall time of 5 ms to avoidaudible clicks from the speaker. The US was a 50-ms train ofconstant current pulses (50 Hz, 1-ms pulse width, 2–3 mA)delivered through the periorbital electrodes. Rabbits were
trained at four different ISIs using the tone CS (128 animals),200, 250, 500, and 1000.
Rabbits receiving daily eyelid conditioning sessions withmossy fiber stimulation as the CS (32 animals) consisted ofpassing cathodal current stimulation (100 Hz, 100-μs pulsewidth, 100 μA) through electrodes implanted in the middlecerebellar peduncle. Rabbits were trained with four differentISIs using mossy fiber stimulation as the CS, 250, 500, 750,and 1500. Both tones and mossy fiber stimulation as CSslasted for 50 ms longer than the ISI to allow the CS and USto co-terminate. Trials using both tones and mossy fiber stim-ulation were separated by a mean intertrial interval of 30±10 s.
Data Analysis
For each trial, 2500 ms of eyelid position (200 ms pre-CS,2300 ms post-CS) was collected at 1 kHz and at 12-bit reso-lution. Data were stored to a computer disk and analyzed off-line using custom-written scripts in MATLAB. Eyelid posi-tion measured 200 ms before each trial established a baselinefor detecting eyelid movement during each trial. Trials wereexcluded from the analysis if a movement greater than 0.3 mmoccurred during the 200-ms baseline period before CS onset.A CR was defined as an eyelid closure of at least 0.3 mmwithan onset between 30 ms after CS onset and the onset of theUS. Amplitude of the CR was defined as the value of eyelidposition at the time of US onset (or where it would had beenfor CS-alone trials). Error bars in all figures indicate SEM. CRpercentage or likelihood, used in the analysis, refers to thefraction of trials where the response satisfied the CR criterion(number of CR trials divided by the total number of validtrials). In the present analysis, we used amplitude and CRlikelihood measures to investigate the single CS-alone trialeffect. For all analysis except acquisition results shown inFig. 1, we included only sessions with at least 70 % CRs,corresponding typically to sessions 3–5 from Fig. 1
Results
All data were taken from animals trained for other purposes,such as acquiring CRs prior to a lesion or reversible inactiva-tion experiment. All of these manipulations occurred after thedata presented here. It is also the case that none of the animalshad received prior training experience at a different ISI. Theanalyses make use of the fact that with a larger than usualnumber of subjects per group, it is possible to detect smallbut significant differences in the amplitudes of CRs in thetrials that follow the CS-alone test trials that are presented aspart of our normal training protocols. Figure 1a shows a pro-totypical acquisition curve for eyelid conditioning, in this in-stance for the animals that were trained for five sessions using
Cerebellum
a tone CS and an ISI of 500 ms. As is not uncommon foreyelid conditioning experiments, these data are expressed asmean amplitude for each block of nine trials, in this instance12 blocks per training session with the five training sessionsindicated by each separation in the graphs. The mean CRamplitude grows over the 5 days of training with characteris-tics that have been observed and analyzed previously. Forexample, after initial training, there is a clear decrease in CRamplitude within sessions (sessions 3–5) that recovers be-tween sessions, consistent with the short-term plasticity pro-cesses that have been shown to operate to decrease CR ampli-tude [15, 16]. Figure 1b shows the same data from approxi-mately half of these animals where the data are expressed asaverages of single trials across the five sessions. These data
are from tone CS and ISI 500 animals that received CS-alonetrials at the beginning of each block; similar results were ob-served from the remaining animals where the CS-alone trialwas at the end of each block (not shown). Expressed as aver-ages of single trials, we can now see that there are regulardecreases in response amplitude during sessions with robustresponding (the last three sessions in this case). These smalleramplitude responses are from the trials that immediatelyfollowed the CS-alone trials and are indicated with gray cir-cles below the average curves in the last three sessions. In theremainder of this manuscript, we analyze this effect in eightdifferent groups of animals: (1) four trained with a tone CSand one of four different ISIs (200, 250, 500, and 1000 ms),and (2) four groups trained with mossy fiber stimulation as theCS, and using one of four different ISIs (250, 500, 750, and1500 ms).
This effect following CS-alone trials is observed moreclearly when data are collapsed across all blocks for all train-ing sessions that satisfied the criterion of greater than 70 %CRs over the session, and then collapsed across all animals inthe same CS/ISI group (Fig. 2). All data points in the upperpanels of Fig. 2 are normalized to the mean CR amplitude ofthe CS-alone response and the responses from the two preced-ing trials. Each graph shows these three trials and the eightpaired CS + US trials that followed the CS-alone trial (indi-cated by the red arrow in each panel). The data for the four ISIgroups trained using a tone CS are shown in Fig. 2a, with theISI groups color-coded as indicated. The upper panel showsthat there was a decrease in mean CR amplitude in the trialfollowing a CS-alone trial and that this effect is clearly largerfor the longer two ISIs (ISI 500 and 1000) that were tested.This same panel also shows that over the subsequent seven CS+ US trials, the CR amplitude gradually increases. The hori-zontal black bar indicates the trials for which there was asignificant overall decrease in CR amplitude as compared tothe CS-alone trial. The asterisk indicates the trials for whichthe decrease for the different ISIs were significantly different.These changes in the mean CR amplitude could potentially beaccounted for entirely by a decrease in CR likelihood. In orderto examine this possibility, we performed the same analysesfor the CR likelihood measure (Fig. 2a, middle panel). Herethe change following the CS-alone trial is clearly smaller thanthe changes in amplitude, suggesting that the effect on ampli-tude is at least not entirely explained by an increase in CRfailure following a CS-alone trial. The histogram at the bottomof Fig. 2a shows for each of the four ISIs the mean change inresponse amplitude (in mm) between trials just preceding andimmediately following the CS-alone trial; a negative valueindicates the response amplitude decreased after the CS-alone trial. The decrease in amplitude is significant for themajority of ISIs (two-tailed t test, p<0.05 for ISI 250 andp<0.01 for ISIs 500 and 1000), and differences between ISIgroups are indicated by horizontal bars (one-way ANOVA,
0
1
2
3
4
5
Mea
n C
R a
mp
(mm
)
Trial number
12 24 36 48 600
1
2
3
4
5M
ean
CR
am
p (m
m)
Block number
108 216 324 432 540
Fig. 1 a The acquisition of conditioned eyelid responses using a 500-msISI is reflected in a systematic increase in mean response amplitude over5 days of training. Each training session was comprised of 12 blocks ofnine trials, with each block comprised of eight CS+US paired trials andone CS-alone test trial. Each data point is the mean of the block averageacross 99 subjects trained in this way. There are characteristic features ofthis behavior reported in previous papers such as the decline in amplitudewithin each session and recovery seen as an increase in amplitude at thebeginning of each session. b The same data plotted as the average of theindividual trials across 39 subjects. This reveals that there is a noticeabledecrease in amplitude in the trials following the CS-alone trials (trialsafter CS-alone trials indicated by gray circles for the last threesessions). These animals were trained with the CS-alone trials at thebeginning of each block, and the remaining subjects, whose data wereomitted in panel (b), were trained with CS-alone trials at the end of eachblock. These animals showed the same trend: a decrease in amplitude inthe trials following a CS-alone test trial (not shown)
Cerebellum
−2 −1 1 2 3 4 5 6 7 80.6
0.7
0.8
0.9
1
1.1
Nor
mal
ized
CR
am
plitu
de
ISI 200ISI = 250ISI = 500ISI = 1000
0.6
0.7
0.8
0.9
1
1.1
−2 −1 1 2 3 4 5 6 7 8
Trial number
Nor
mal
ized
CR
like
lihoo
d
CS-Alone
0.6
0.7
0.8
0.9
1
1.1
Nor
mal
ized
CR
am
plitu
de
0.6
0.7
0.8
0.9
1
1.1
Nor
mal
ized
CR
like
lihoo
d −2 −1 1 2 3 4 5 6 7 8
Trial number
−2 −1 1 2 3 4 5 6 7 8
CS-Alone
ISI 250ISI = 500ISI = 750ISI = 1500
−0.8
−0.6
−0.4
−0.2
0
CR
am
plitu
de d
ecre
ase
(mm
)
−0.8
−0.6
−0.4
−0.2
0
ISI 200
ISI = 250
ISI = 500
ISI = 1000
ISI 250
ISI = 500
ISI = 750
ISI = 1500
SC rebif yssoM )BSC enoT )Ap < 0.05 p < 0.05
* **
Fig. 2 The decrease in response amplitude following CS-alone trials isseen in animals trained using a tone CS (a) and with animals trained usingmossy fiber stimulation as the CS (b). The upper panels show the effectsof CS-alone trials on CR amplitude by averaging across all CS-alonetrials for all animals. Each data point is normalized to the averageamplitude (for that block, see the BMethods^ section) of the CS-alonetrial and the two trials that preceded it. The red arrow indicates the CS-alone trials. Each graph therefore shows the amplitude of the CS-alonetrial, the two trials that preceded it, and the eight trials that followed it.There are small differences in trials -2 and -1 versus 7 and 8 because, foreach session, either the first CS-alone trial was omitted if it was the veryfirst of the session or the last CS-alone trial was omitted if it was the verylast trial of the session. The middle panels show CR likelihood datanormalized in the same fashion. The bottom graphs show for thedifferent ISI groups the mean change of CR amplitude (in mm) in thetrial that followed a CS-alone trial compared to the trial preceding it. aUsing a tone CS, there was a significant decrease in CR amplitude after aCS-alone trial (upper panel) that (1) recovered over the next eight pairedCS+US trials and (2) generally increased in amplitude as the ISIincreased. Four ISIs were used with a tone CS, 200 ms (gray), 250 ms
(black), 500 ms (red), and 1000 ms (blue). The middle panel shows thateffects are smaller with the CR likelihood measure, indicating that thedecrease in CR amplitude following CS-alone trials is not entirelyexplained by an increase in CR failures. The bottom panel shows thatthere were significant differences for each ISI and between the ISIs asindicated by the horizontal bars and asterisks. b The same general resultswere seen in animals trained with mossy fiber stimulation as the CS ratherthan a tone CS. The upper panel shows the normalized amplitude data,with clear decreases in the amplitude of the trials following the CS-alonetrial. These decreases were larger for longer ISIs and recovered over thefollowing eight CS+US paired trials. Themiddle panel shows, as with thetone CS, that the effects were much smaller when expressed asnormalized CR likelihood. The bar graph at the bottom shows againthat the mean decrease in amplitude is significant for each ISI andincreases across the ISIs. Significant differences between ISI groups areshownwith horizontal bars and asterisks. As with the tone CS data, blackindicates ISI 250 ms and red indicates ISI=500 ms. With mossy fiberstimulation animals, there were two new ISIs, green indicates ISI=750 ms, and cyan indicates ISI=1500 ms (Color figure online)
Cerebellum
F(3,369)=5.19, p=0.002). The four groups trained using mossyfiber stimulation as the CS showed essentially the same pat-tern of results (Fig. 2b). The top panel of Fig. 2b shows asimilar decrease in CR amplitude following CS-alone trialsas was seen with the tone CS. These results are also similarin that the largest effects on amplitude were seen with thelonger ISIs. Also similar to the tone CS data, there was a muchsmaller effect observed in the CR likelihood measure (middlepanel, Fig. 2b). The histogram at the bottom of Fig. 2b com-pares the decrease in CR amplitude across the four ISIs tested.ISIs are color-coded, and ISIs tested for both tone and mossyfiber CSs maintain color coding (black = ISI 250, red = ISI500 ms). As with tone, CR amplitude decreases were signifi-cant for the majority of ISIs (two-tailed t test, p<0.01 for ISIs500, 750, and 1500), and the larger decreases in CR amplitudewere observed with the longer ISIs (one-way ANOVA, F(3,57)=4.11, p=0.011)
Together, these data show statistically reliable decreases inCR amplitude following CS-alone trials. We observed thesame pattern independent of whether the CS was a tone ormossy fiber stimulation, eliminating the possibility that thecause of the single-trial effect can be upstream of the cerebel-lum. This effect increases with the ISI, consistent with obser-vations that the rate of extinction in delay eyelid conditioningis quite slow for shorter ISIs and is much faster for longer ISIs(Ohyama and Mauk, unpublished). The incremental nature ofcerebellar learning is illustrated in Fig. 2, which shows howthe mean CR amplitude or likelihood returns to the baselinelevel with additional CS + US trials. While these increases aregenerally consistent with the Bclimbing fiber is present^ por-tion of the Medina and Lisberger study, we do not pursue itfurther here because we have no way at the level of behaviorto parse trials where there was a complex spike present fromthose where it was absent. Overall, this pattern of results isconsistent with the Bclimbing fiber absent^ portion of the Me-dina and Lisberger observation of single-trial changes in be-havior, with the addition that, in this case, these changes cor-relate with the variations in the rate of learning (extinction)over different experimental conditions.
To characterize the nature of the changes in CRs producedby single CS-alone trials, we calculated frequency histogramsof the differences between CR amplitudes in the trials justpreceding and immediately following CS-alone trials(Fig. 3a, b). Here a negative value indicates a decrease inamplitude compared to the preceding CS-alone trial. The his-tograms are arranged with ISI increasing from left to right andare staggered so that the two ISIs used for both mossy fiberstimulation (Fig. 3a) and tone CS (Fig. 3b) animals are alignedvertically (ISI=250 ms shown in the gray box and ISI=500 inthe lighter gray box). For each of the eight panels in Fig. 3a, b,zero change is denoted with a vertical dotted line and also ineach panel the mean (black arrow) and the median (gray ar-row) are less than zero, indicating a decrease in response
amplitude. Distributions for each ISI and CS are skewed inthe negative direction reflecting that the decrease in CR am-plitude after CS-alone trials is a general property of cerebellarlearning.
Because the trial-to-trial variability of CR amplitude in-creases with the ISI, we compared the changes in CR ampli-tude following a CS-alone trial with the population of trial-to-trial changes in CR amplitude for each ISI. Figure 3c showsfrequency histogram of the population trial-to-trial change inCR amplitude. These were calculated by randomly samplingpairs of trials that a trial in between them from the same ses-sions used to calculate the effects of CS-alone trials. Figure 3dshows the differences between the population distributionsand the CS-alone distributions for five groups. Each ISI in-volved in this population analysis is shown with a uniquecolor other than black (cyan = ISI 200, green = ISI 250, blue= ISI 500, gray = ISI 1000, and red = ISI 500 with mossy fiberstimulation as the CS). These difference distributions (CS dis-tribution minus population distribution) show that for allgroups, except ISI 200, trials following CS-alone trials dem-onstrate more frequent decreases in CR amplitude and fewerincreases in CR amplitude than would be expected from thepopulation changes. Moreover, the panels in Fig. 3d show thatthe likelihood of larger decreases in CR amplitude increasewith the ISI.
Next, we asked how CS-alone trials affect the time courseor temporal profile of the CRs at each ISI. Figure 4 showsaverage response profiles averaged across the four ISIs usedin tone training (Fig. 4a) and the four ISIs used in trainingwhere the CS was stimulation of mossy fibers (Fig. 4b). Thetop panel in Fig. 4a shows average response profiles for theCS-alone trials (black traces) and for the paired CS + US trialsthat preceded them (green traces). Four pairs of traces areshown, one each for the four ISIs as indicated. These tracesshow that the CS-alone trials and those that precede them arealmost indistinguishable through the whole eyelid trajectoryprior to the US, having the same CR onset and amplitude. Theshaded region of each trace indicates 95 % confidence inter-vals. The only notable difference in the four pairs of traces isthe reflex response to the US that occurs on the CS + USpaired trials. The reflex response for ISI 200 has been omittedto avoid overlap with the ISI 250 traces. In contrast, the lowerpanel of Fig. 4a shows the small but significant differencesbetween the CS-alone trials (black traces) and the trials thatimmediately followed (red traces). At the longer two ISIs,there are clear differences in the time profiles and subsequentamplitude of the CRs, while the onset of the responses remainconsistent. The same data, with essentially the same pattern ofresults, from animals trained with mossy fiber stimulation asthe CS are shown in Fig. 4b.
The mean amplitude of CRs on the trial after the CS-alonetrial shows a consistent decrease that then recovers to normallevels during the subsequent eight paired CS + US trials. Still,
Cerebellum
looking only at the mean values of CR amplitude conceals thequestion of how the entire distribution of CRs is influenced bythe CS-alone trial. To address this question, we used heatmaps to illustrate how the distribution of CR amplitudeschanges after the CS-alone trial (Fig. 5). For all ISIs and bothtone and mossy fiber stimulation CSs, we calculated the mean
distribution of CR amplitudes on CS-alone trials and the twotrials preceding it and plotted changes from that baseline dis-tribution on trials following the CS-alone trial. The single-trialeffect on the distribution of CR amplitudes is not clear at shortISIs, but starting from ISI 500, a common trend emerges. Forboth types of CSs, following the CS-alone trial, there is an
.02
.04
.06
.08.1
ISI 1000
.02
.04
.06
.08.1
.12
.14
ISI 500
.04
.08
.12
.16
.2
ISI 250
B)TONE CS
.04
.08
.12
.16 ISI 200
Pro
port
ion
of r
espo
nses
.02
.04
.06
.08.1
.12
.14
.16ISI 200
.02
.04
.06
.08.1
.12ISI 500 (MF)
Pro
port
ion
of r
espo
nses
-.04
-.02
.02
.04
0
-.04
-.02
.02
.04
0
-.04
-.02
.02
.04
0
-.02
-.01
.01
.02
0
-.02
-.01
.01
.02
0
ISI 250
.02
.04
.06
.08.1
.12
.14
.16ISI 500
.02
.04
.06
.08.1
.12
.14
.16ISI 1000
.02
.04
.06
.08.1
.12
.14
.16
Diff
eren
ce o
f Pro
port
ions
−6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6
−6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6
−6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6
Trial to Trial Change in CR Amplitude (mm)
Trial to Trial Change in CR Amplitude (mm)
−6 −4 −2 0 2 4 6
C) DISTRIBUTION OF ALL TRIAL TO TRIAL CHANGES
Pro
port
ion
of r
espo
nsesA) MF STIMULATION
.02
.04
.06
.08 ISI 1500
.02
.04
.06
.08 ISI 750
.02
.04
.06
.08
.1
.04
.08
.12
.16
.2
−6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6 −6 −4 −2 0 2 4 6
D) RELATIVE TO ALL TRIAL TO TRIAL CHANGES
Fig. 3 Changes in CR amplitude following CS-alone trials are expressedas frequency histograms of the difference between the amplitude of CRsjust preceding and immediately following CS-alone trials—negativevalues indicate smaller responses on the trial following the CS-alonetrial. Separate histograms are shown for each ISI with data fromanimals using mossy fiber stimulation as the CS along the top row (a)and data from animals trained with the tone CS along the second row b. Inpanels (a) and (b), the histograms are arranged with ISI increasing fromleft to right and are staggered so that the two ISIs used for both tone andmossy fiber stimulation animals are aligned vertically (ISI=250 msshown in the gray box and ISI=500 in the lighter gray box). For eachof the eight panels, zero change is denoted with a vertical dotted line andalso in each panel the mean (black arrow) and the median (gray arrow)are less than zero, indicating a decrease in response amplitude. c, d Thechanges in CR amplitude following CS-alone trials can be betterunderstood when compared to the typical change in CR amplitude fortrial pairs of randomly selected trials that have a trial in between them (tomatch the comparison, a trial following CS-alone versus the trial prior tothe CS-alone trial). c These population frequency histograms wereconstructed for five groups, four tone-CS groups (ISI 200=cyan, ISI250=green, ISI 500=blue, and ISI 1000=gray), and the one mossy
fiber-CS group for which there was enough data to perform thisanalysis (ISI 500=red). These five histograms were constructed byrandomly selecting with replacement a total of 10,000 pairs of trialsfrom the same sessions used for panels (a) and (b). These figures revealthe unsurprising result that there is more variability in CR amplitude forlonger ISIs than for shorter ISIs. d To compare the changes in CRamplitude after CS-alone trials with the typical (population) variability,the population frequency histograms in (c) were subtracted from therelevant (color-coded) frequency histograms in (a) and (b). Forexample, the red histogram in panel (c) was subtracted from the redhistogram in panel (a) to produce the red difference histogram in panel(d), which is for the mossy fiber stimulation CS and ISI=500 the changein CR amplitude produced by CS-alone trials normalized to the typicalchanges between trials that is observed in the same sessions. In general,these difference histograms show that CR amplitude decreasesmore oftenthan expected after CS-alone trials and increases less often than expected.The results for ISI 200 are the exception, where it is difficult to see asignificant trend. Comparing across ISIs, these difference histogramsshow that there tends to be small decreases in amplitude at shorter ISIsand larger decreases in amplitude for longer ISIs (Color figure online)
Cerebellum
increase in the frequency of smaller amplitude responses and adecrease in the frequency of large amplitude responses. As thelength of the ISI increases, this effect becomes especiallystrong and persists over several trials following the CS-alonetrial. These plots demonstrate how specifically CS-alone trialsdecrease the mean amplitude of CRs in trials immediatelyafter them and how the distribution of CR amplitudes recoversto the baseline level as paired CS + US training within theblock of nine trials accumulates.
Discussion
The current study is in part a replication of previous findingsinitially reported from the smooth pursuit experiments [1, 2],which described climbing-fiber-dependent changes inPurkinje cell activity and behavior on a trial-by-trial basis.Here, our proxy for recording climbing fiber input was to lookat the behavioral effects of CS-alone trials, since both theoret-ical models predicted [17] and recording studies [6] (unpub-lished observations from our lab) have shown a strong
decrease in the likelihood of a climbing fiber input to Purkinjecells during early CS-alone trials in extinction training. Re-sults of the current study extend previous findings by describ-ing how the single-trial effect interacts with a cerebellum-dependent behavior over a range of differently timed CRstrained at various ISIs. The current experiment also isolatedthe decrease in CR amplitude due to the single CS-alone trialeffect to mechanisms inside the cerebellum through use ofmossy fiber stimulation as the CS.
The main finding of this analysis is consistent with previ-ous studies and demonstrates a statistically reliable decrease inanimal performance after a single CS-alone trial, again sug-gesting the incremental nature of cerebellar learning. Havingthe data collected from a large number of animals trained atdifferent ISIs allowed us to examine this effect in detail. Gen-erally, we did not see any reliable differences in the single-trialeffect between groups of animals trained with either tone orelectrical mossy fiber stimulation as the CS. Some differencesin the mean value of CR amplitude reduction or the signifi-cance level of effects between identical ISIs (Fig. 2, ISIs 250and 500) for the two CS types were likely due to a generally
200 400 600 800 10000
1
2
3
4
5
6
Time after CS onset (ms)
Eye
lid p
ositi
on (
mm
)
ISI 200
ISI 250ISI 500
ISI 1000
200 400 600 800 1000 1200 1400 1600
1
2
3
4
5
6
ISI 250
Time after CS onset (ms)
ISI 500
ISI 750
ISI 1500
200 400 600 800 10000
1
2
3
4
5
6
Time after CS onset (ms)
Eye
lid p
ositi
on (
mm
) ISI 1000
ISI 200
ISI 250 ISI 500
1200 1400 200 400 600 800 1000 1200 1400 1600
1
2
3
4
5
6
ISI 250
Time after CS onset (ms)
ISI 500
ISI 750
ISI 1500
A) Tone CS B) Mossy fiber CS
CS CS
CS CS
Fig. 4 The decrease in CR amplitude seen in the trials following CS-alone trials is also seen in averaged response profiles. For each panel, theabscissa shows time in milliseconds following CS onset (indicated byarrow) and the ordinate is mean CR amplitude in millimeters. Eachsweep in these four panels is the average response profile over allresponses across all animals for the ISI indicated. In all cases, theshaded portion of the sweep depicts the 95 % confidence interval forthe data. The top panels compare for each ISI the average responseprofile of the CS-alone trials (black traces) and the trials preceding eachCS-alone trial (green traces). The bottom two panels compare for each ISIthe average response profile for the CS-alone trials (again in black)compared to the trials following each CS-alone trial (red traces). a Theupper panel shows that CS-alone trials were almost indistinguishablefrom the responses in the paired trials that preceded them. The only
regions of non-overlap in the responses are the reflex responses to theUS that occur on the paired trials (the reflex response for ISI 200 wasomitted to avoid overlap with the responses from ISI 250). These dataalso clearly show the robust timing differences between the responses atdifferent ISIs. Only the mean responses for ISIs 200 and 250 ms are notobviously different. The bottom panel in contrast shows small butsignificant decreases in amplitude of CRs that follow CS-alone trials,especially at the longer two ISIs. b The same general pattern of resultswas seen from animals where the CS was stimulation of mossy fibers.The upper panel shows the similarity between CS-alone trials and thetrials that precede them whereas the lower panel shows clear decreases inthe amplitude of trials that follow CS-alone trials. As with the tone CSdata, the decrease in response amplitude was larger for the longer ISIs(Color figure online)
Cerebellum
larger number of animals trained with the tone CS. Thus, webelieve that at least the main source of the single-trial effect isconstrained to learning mechanisms inside the cerebellum,which is fully consistent with reported changes in Purkinjecell firing determined by the presence or absence of climbingfiber inputs from Lisberger’s lab.
We observed that a single CS-alone trial affects not onlyCR amplitude on the following trial but also its whole averagetime profile (Fig. 4). It also alters the distribution of CR am-plitudes by decreasing the frequency of CRs with the highestamplitudes and increasing the frequency of both non-CRs and
low amplitude CRs (Fig. 5). The distribution of intermediateamplitude CRs, on the other hand, did not seem to be influ-enced by the CS-alone trials. That suggests a specific way inwhich the cerebellum is implementing a decrease in averageCR amplitude: It is not the case that the whole distributionbecomes skewed to change the mean CR amplitude value.Rather, the majority of the distribution of CR amplitudes ispreserved with only the frequencies of the lowest and thehighest amplitudes changing.
These changes in amplitude distributions for each ISI couldarise in two general ways: There could be a tendency for large
% change of distribution
−10 −5 0 5 10 15
Trial number−2 0 2 4 6 8
Trial number−2 0 2 4 6 8
−10
−5
0
5
10
Bin
ned
CR
Am
plitu
de (
mm
)
% change of distribution
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
−10 −5 0 5 10 15
A) Tone CS B) Mossy fiber CS
ISI 200
ISI 250
ISI 500
ISI 1000
ISI 250
ISI 500
ISI 750
ISI 1500
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
Fig. 5 Color-coded heat maps and histograms illustrate the changes inresponses of various amplitudes over the eight trials following a CS-alonetrial. Data for tone CS is shown at left (a) and data for mossy fiberstimulation is shown at right (b). The amplitudes of all responses werebinned (0.5-mm bin width), with bins ranging from 0 to .5 mm at thebottom of each graph and 5.5–6.0 mm along the top. Each graphrepresents with color the percent change of each bin of the CRamplitude distribution, with negative values denoting a decrease in thenumber of responses at that amplitude. Each heat map shows data from adifferent ISI, with ISIs increasing from top to bottom. The abscissa foreach heat map shows the CS-alone trials (the bin to the left of the blackvertical lines), the two trials that precede the CS-alone trial (far left twobins), and the eight trials that follow the CS-alone trials (all bins to theright of the black lines). The bar graphs to the left of the tone CS heat
maps and the right of the mossy fiber CS heat maps show, with the samecolor coding used in the heat maps, the change in the CR amplitudedistributions for the trial immediately following the CS-alone trial(notice that the colors of each bar is the same as the correspondingamplitude bin just to the right of the black line in each heat map). Theheat map colors to the left of the vertical bars are generally neutral colors,showing that the responses have returned to their usual amplitudes. Incontrast, the bins just to the right of the vertical bar tend to show coolercolors in the larger amplitude bins and warmer colors in the smalleramplitude bins, indicating a decrease in response amplitude after theCS-alone trials. The heat maps demonstrate, for each ISI, how the CRamplitude distribution returns to the baseline level in the trials that followthe CS-alone trial. The color calibration bar shows the percent changefrom the full distribution of response amplitudes (Color figure online)
Cerebellum
responses to become non-CRs, or there could be a tendencyfor each response to decrease by a small amount. Figure 3suggests that for shorter ISIs, the latter is more applicable, asthe modal change is a small decrease in amplitude (and thedistribution is relatively narrow). For these shorter ISIs, Figs. 3and 5 show that most CS-alone trials produce a small decreasein CR amplitude of the next trial. However, results from thesame two figures suggest that for longer ISIs, the situation isslightly different: There, decreases in CR amplitude followinga CS-alone trial are larger and there is a tendency for morenon-CRs. In both cases, this is generally consistent with casualobservation of response changes during extinction. For shorterISIs, we generally observe a steady decline in CR amplitudeover many trials, whereas for longer ISIs, we observe de-creases in amplitude that often result in the abrupt disappear-ance of the CR. In ongoing work, we hope to understandbetter the mechanisms that contribute to these differences.
Since climbing fiber suppression is the signal for extinctionin eyelid conditioning [13], the single-trial effects observed inthe current study are likely due to a small amount of extinctionduring CS-alone trials. In delay eyelid conditioning, there is astrong effect of ISI on the rate of extinction (Ohyama andMauk, unpublished) such that extinction is quite slow forshorter ISIs and much faster for longer ISIs. For short ISIs(e.g., ISI 250) animals will not extinguish for hundreds oftrials while extinction at longer ISIs (e.g. ISI 750) is often veryrapid, occurring in around 20 trials, for both tone and mossyfiber stimulation conditioned animals. Consistent with the re-lationship between extinction and ISI, all of the effects wereport here also significantly scale up with the length of theISI. If the cerebellar mechanism for extinction and the single-trial effects observed in the current analysis are the same, theimplication based on the current results would be that behav-ioral changes produced by cerebellar learning during extinc-tion become faster as the ISI increases.
The current results also show that not only is the magnitudeof the single-trial effect larger at the longest ISIs but also thatmore paired CS + US trials are needed for CRs to recover tofull amplitude at long ISIs as well (Fig. 2, upper panels). It iswell established that the initial rate of learning decreases withlonger ISIs as animals transition from the naïve state to thelearned state in eyelid conditioning [10, 16]. On the otherhand, rate of extinction and changes due to single CS-alonetrial effects scale in the opposite way by increasingwith longerISIs. That suggests that cerebellar learning mechanisms arearranged in a way that makes it harder to learn and faster toextinguish CRs as the ISI increases. In future studies, thecerebellar mechanisms related to the scaling rate of the singleCS-alone trial effect, during extinction and acquisition withdifferent ISIs, will need to be examined with cerebellar simu-lations and recordings from Purkinje cells during eyelid con-ditioning. It will be important, for example, to identify wheth-er the cerebellar mechanisms underlying the single-trial
effects observed during eyelid conditioning are similar to re-sults observed during smooth pursuit learning [1, 2].
Acknowledgments This work was supported byNIH grants MH74006and MH46904.
References
1. Medina JF, Lisberger SG. Links from complex spikes to local plas-ticity and motor learning in the cerebellum of awake-behavingmonkeys. Nat Neurosci. 2008;11:1185–92.
2. Yang Y, Lisberger SG. Purkinje-cell plasticity and cerebellar motorlearning are graded by complex-spike duration. Nature. 2014;510:529–32.
3. Mauk MD, Steinmtez JE, Thompson RF. Classical conditioningusing stimulation of the inferior olive as the unconditioned stimu-lus. Proc Natl Acad Sci. 1986;83:5349–53.
4. Steinmetz JE, Lavond DG, ThompsonRF. Classical conditioning inrabbits using oointine nucleus stimulation as a conditioned stimulusand inferior olive stimulation as a n unconditioned stimulus.Synapse. 1989;3:225–33.
5. McCormick DA, Thompson RF. Neuronal responses of the rabbitcerebellum during acquisition and performance of a classically con-ditioned nictitating membrane-eyelid response. J Neurosci. 1984;4:2811–22.
6. Gould TJ, Steinmetz JE. Changes in rabbit cerebellar cortical andinterpositus nucleus activity during acquisition, extinction, andbackward classical eyelid conditioning. Neurobiol Learn Mem.1996;65:17–34.
7. Halverson HE, Lee I, Freeman JH. Associative plasticity in themedial auditory thalamus and cerebellar interpositus nucleus duringeyeblink conditioning. J Neurosci. 2010;30:8787–96.
8. Svensson P, Ivarsson M, Hesslow G. Effect of varying the intensityand train frequency of forelimb and cerebellar mossy fiber condi-tioned stimuli on the latency of conditioned eye-blink responses indecerebrate ferrets. Learn Mem. 1997;4:105–15.
9. Kalmbach BE, Ohyama T, Mauk MD. Temporal patterns of inputsto cerebellum necessary and sufficient for trace eyelid conditioning.J Neurophysiol. 2010;104:627–40.
10. Schneiderman N, Fuentes I, Gormezano I. Acquisition and extinc-tion of the classically conditioned eyelid response in the albinorabbit. Science. 1962;136:650–2.
11. Sears LL, Steinmetz JE. Dorsal accessory inferior olive activitydiminishes during acquisition of the rabbit classically conditionedeyelid response. Brain Res. 1991;545:114–22.
12. Hesslow G, Ivarsson M. Inhibition of the inferior olive during con-ditioned responses in the decerebrate ferret. Exp Brain Res.1996;110:36–46.
13. Medina JF, NoresWL, MaukMD. Inhibition of climbing fibres is asignal for the extinction of conditioned eyelid responses. Nature.2002;416:330–3.
14. Ryan SB, Detwiler KL, Holland KH, Hord MA, Bracha V. A long-range, wide field-of-view infrared eyeblink detector. J NeurosciMethods. 2006;15:74–82.
15. Kreider JC, Mauk MD. Eyelid conditioning to a target amplitude:adding how much to whether and when. J Neurosci. 2010;30:14145–52.
16. Ohyama T, Voicu H, Kalmbach B, Mauk MD. A decrementingform of plasticity apparent in cerebellar learning. J Neurosci.2010;30:16993–7003.
17. Mauk MD, Donegan NH. A model of Pavlovian eyelid condition-ing based on the synaptic organization of the cerebellum. LearnMem. 1997;4:130–58.
Cerebellum